ACWWong
|
Hi Aaron,
the system design should clearly give you specifications of your filter, and these specifications will determine whether the design is difficult or easy. Specifications such as filter order & shape (i.e pole/zero positions) are determined from reciever adjacent channel rejection and inband ripple & groupdelay (requirements depends on the data modulation type). Having been given the filter order & shape, then one usually chooses and filter architecture (Ladder, cascaded biquads, SFG etc.) in conjucntion with realization type (opamp RC, gmC etc.) ANyway, There is alot of stuff on the web about gmC (OTA-C) filters, try circuitsage.com or anything by rolf schaumann for basic information. Also the book by Laker & Sansen is good starter.
In answer to your direct questions: Where in opampRC designs one uses R*C to produce the timeconstants required for the transfer function, OTA-C uses gm/C. so yes its the value of the OTA's gm which is used. so for a pole/zero frequency~ 3.5MHz, with C's about 10pF we are talking ~220uS. The same timeconstant could be used with C=20pF and 440uS for lower noise but at the cost of area/power. The bandwidth of the OTA will depend on the Q and frequency of the filter poles you are trying to realise and what and where they are used in your filter architecture. Generally you'll be pretty safe with Q^2*Fc for the OTA bandwidth. The other important thing for gmC filters is linearity, as often the bottleneck for linearity in buried within the filter architecture (usually where the Q is highest) and often the OTA linearity requirement is much larger than the overall filter linearity spec.
Practical considerations such as CMFB are you state are always inportant (there's nothing worse than an oscillating CM loop!), but PSRR and CMRR are usually ok if one uses differential toplogies.
sorry for my long winded response, as you can see continuous time filter design is a large subject!
cheers aw
|